Speak EV - Electric Car Forums banner

1 - 17 of 17 Posts

·
Banned
Joined
·
35 Posts
Discussion Starter #4
RIGHT THEN!
mythz2.png


According to that graph the crash rate before Autopilot was 1.3 per million miles and after Autopilot was 0.8. At the time the report came out there were approximately 100,000 Teslas on the road in USA. Average driver does about 10,000 miles a year so the total mileage each year was about 1000 million miles.

So there were 1300 Tesla crashes that were bad enough to deploy airbags each year before Autopilot and a mere 800 crashes a year afterwards,

HOLY CRAP!!!

That's UNBELIEVABLE

(but even then it's too small a sample to prove the 40% safer claim)

Also bear in mind that by looking at all the salvage Teslas in the junkyard it is certain that every crash bad enough to deploy the air-bags means the car is beyond economic repair and will be an insurance write off.

I say... MYTH BUSTED!
 

·
Banned
Joined
·
35 Posts
Discussion Starter #5
Tesla provided the Office of Defects Investigation with mileage and airbag deployment data for all 2014-2016 Model S and Model X vehicles equipped with Autopilot technology. The investigators used this data to calculate crash rates prior to and after Autopilot installation.

Oh REALLY?
 

·
I'm not crazy, the attack has begun.
Joined
·
23,602 Posts
I've found UK 2011 statistics
100 billion car and taxi miles (Gov data) for 2.2 million accidents (RAC data)
which makes 22 accidents per million miles, so Tesla's 1.3 accidents per million miles looks pretty good to me as it was.

TBH the rate is so low without autosteer that the statistics are more Poisson than normal, so the 40% decrease is into statistical insignificance, and therefore not arithmetically surprising.

So factually the statements appear correct, only due to statistical mathematics might they be dubious. But that is not something that would concern a court, the case can be dismissed.
 

·
Banned
Joined
·
35 Posts
Discussion Starter #10
Apples/oranges?

Tesla stats are for crashes serious enough to cause air bag deployment.
Not just fender benders.
Your stats include old cars, inexperienced and impaired drivers, fraudulent insurance staged accidents, stolen cars.
Need to compare data for similar large luxury cars owned (mostly) by mature experienced drivers.

Even then, 1300 serious crashes in Tesla is a very high number.
But not anywhere near enough to prove a 40% crash reduction with any statistical reliability.
 

·
Registered
Joined
·
932 Posts
I do this type of thing for a living (crunching numbers) and you can make anything look like anything. I'm not saying its right or wrong, just you pick the numbers that suit your agenda.

As an example - is the accident rate with AP enabled or not? If not then what difference does it make?

If its AP driven miles, then its often turned off for the trickier bits of the road which is when accidents are more likely. The empty motorway on AP is almost certainly always going to be a really safe way to drive.

What about EAP and the long period when it could only be used at slower speeds. AP also changes driver behaviour - I find I'll stick it on and do 70mph, without my speed may be different. Speed is a contributor to accidents. My personal driving style would differ when using AP or not - while thats still a reduction in accidents is it AP thats doing it or is it AP thats making me do it (although we could argue it doesn't matter or we could argue as driver become more familiar with it, then poor driving standards will increase)

My hunch is AP means we drive differently, I can't believe AP in itself is accountable for increasing safety. I'd love to be proven wrong
 

·
Registered
Joined
·
2,857 Posts
If its AP driven miles, then its often turned off for the trickier bits of the road which is when accidents are more likely. The empty motorway on AP is almost certainly always going to be a really safe way to drive.
The original comparison (made in the NHTSA report and quoted by Tesla) was for all miles driven, and for (mostly) the same set of cars, before and after Autopilot was enabled. So the comparison was between miles driven with autopilot available (whether the driver chose to use it or not) against miles driven with autopilot unavailable.

So that's a pretty good comparison so far as it goes, but as you point out there are other factors such as changing driver behaviour over time and changes to the autopilot itself - is today's autopilot better or worse than the one for which statistics are available?

There's also a public perception issue which applies to both this current autopilot debate and also to full AV (and indeed many other safety systems outside cars): the AP-equipped cars will have different accidents to the conventional cars. There will inevitably be accidents with AP (or FSD) where you would say "this accident would almost certainly not have happened with a human driver", and meantime there will be accidents with human driving where you can say "AP would have avoided this accident". Tesla is arguing that an overall lower accident rate is better. I agree, but the public doesn't always: we currently enforce a safety level at least 5x greater on railways than on cars, to the extent that the greater costs force people onto the roads who might otherwise have taken the train, because rail accidents tend to be huge and shocking (while very rare) compared to road accidents that happen every day and don't deliver the same impact that "something must be done".
 

·
Registered
Joined
·
958 Posts
Also do they include the collision warning as part of auto pilot?

That definitely saved my bacon the other day.

In my Leaf or Prius I would definitely have made a car shaped hole in their rear bumper!
 

·
Registered
Joined
·
932 Posts
Its an interesting debate which will run several times I'm sure. I think we've even had the debate on here before.

Its hard to see how an inactive device can be that much safer than a car without the inactive device - so it may be linked to the always on safety features but they're far from new or unique and arguably not what AP is all about. They're also not linked to the enabling of the AP option - they work if you don't tick the box so in many respect the argument should be current Teslas safety kit and not AP.

I think the point about different accidents is pretty valid although I suspect the bar for a safety system will be significantly higher than for people operated stuff. Its simply not generally acceptable for a computerised thing to be in charge when people get injured. I can't think of any examples where that might be the case in anything. I suspect a car crash where the other car is clearly at fault will be excusable, the woman crossing the road at an unfortunate moment and killed by a self driving car - was it reasonable for the car to deal with all such circumstances - no, it had no chance - in much the same way trains aren't all stopped when somebody steps of a railway platform. But we live in a world of lawyers and claims, and saying you were injured because of an AP failure but statistically its meant two other people who we'll never know didn't get hurt through a different scenario doesn't count for much.
 

·
Registered
Joined
·
2,703 Posts
I saw the other day that the XC90 was considered very safe since no one had been killed in one since 2004 in the UK and some wider stats around this and other models.

While I do not doubt the XC90 is a well engineered car, I think the safety record has as much to do with it being in accident where the other vehicles have come off much worse due to the inherent mismatch of mass. So while I don’t doubt there has no one been killed in an XC90 I am pretty sure many many people have been killed being in a collision with one.

How you choose your stats and also state your position can be biased hugely, and to your point, we tend to be more interested in what will be the impact on us, so may choose to drive an inherently safe vehicle. Which in turn might wipe out a Fiesta or 208, does that mean we would choose to be less safe, probably not but it will certainly make a bigger dent. Then the question of fault comes into it, is it the technology, the driver or a 3rd party who is wholly to blame, there isn’t enough granularity in the statistics to allow for that kind of statistical certainty which is why we end up with very blunt stats citing 40% as it is the product of a couple of numbers in a simple calculation.

I kind of believe it as meaningless at this time, the sample needs to be bigger and more insight into the nature of the incidents to make reasoned inferences from it.
 
1 - 17 of 17 Posts
Top